NAOMS Devlog

Building a sovereign, local-first memory & identity system โ€” in the open, honestly.

Fellow Travelers: Reticulum, a network that needs nothing

An encrypted mesh that runs over radio at a few bits per second, with no DNS and no central authority โ€” and what it taught us

Vision Fellow Travelers free May 18, 2026ยท5 min readยทmeta
TL;DR A generous look at Reticulum โ€” an encrypted mesh that keeps working over radio, a serial modem, or a few bits per second, with no central authority and no way to even make an unencrypted link. Here's what its offline-first discipline taught us about keeping a community alive when the internet isn't.

This is not a competitive blog. We build NAOMS inside a generous neighborhood of people doing genuinely hard, principled work, and we study them closely because it makes us better. Few projects embody "works without anything" as completely as Reticulum โ€” a networking stack that assumes almost nothing about the world it runs in, and still delivers an encrypted, identity-proven link.

What they do well

Reticulum runs over whatever medium you happen to have: long-range LoRa radio, packet radio over the airwaves, a serial modem, ordinary internet, WiFi, Ethernet. The same addresses and the same link behavior work across that whole patchwork of wildly mismatched connections. And it works at the very bottom of the bandwidth barrel โ€” reliably at transmission rates as low as five bits per second, with keepalives costing under half a bit per second per link. There is no other way to say it: that is a serious engineering achievement aimed squarely at the places most networks give up.

The design philosophy is sovereign communication. There is no central addressing authority, no global lookup table, no central trust root, and โ€” the part we love most โ€” no way to make an unencrypted link at all. A destination simply drops plaintext as invalid. Encryption is not a setting you can forget to turn on; it is the only thing the protocol knows how to do. Addresses are derived from the hash of a public key, and identity is proven cryptographically at the link layer rather than asserted by an address.

There is real subtlety in the privacy model, too. Packets carry no source address. When you open a link, the destination proves who it is, but you โ€” the initiator โ€” stay anonymous to the network and to every relay in between; only the destination can choose, at the application layer, to learn who reached out. A whole class of metadata leakage that the rest of us treat as a fact of life simply does not exist here. And because no single node holds a complete path โ€” each relay knows only the next hop toward a destination โ€” the network converges on its own, re-finding routes about a minute after the world reappears.

The piece closest to our own concerns is LXMF, a messaging layer built on top. Its propagation nodes are store-and-forward relays: if the person you are messaging is offline or unreachable, any willing node simply holds the message until they come back โ€” a partition measured in days is just a longer wait, not a failure. A single sender-signed envelope carries everything; the envelope is the protocol. That is offline-first not as a fallback, but as the default assumption.

Where to find it

Reticulum is at reticulum.network, with a thorough manual at reticulum.network/manual and a reference implementation on GitHub. It has been in active development since 2016 and runs real off-grid and emergency-communication networks today, with applications like Nomad Network and Sideband built on it. If you want to feel what truly infrastructure-free networking is like, this is where to look.

What we took

We took patterns, deliberately, rather than the whole stack.

The first is the propagation-node discipline for delivering a message to someone who is offline. A single sender-signed envelope, any willing relay can hold it, the recipient pulls it from any reachable node when they return, and no relay is privileged. We already route around offline peers through community mediators, but naming the pattern and checking our design against Reticulum's constraints was a cheap, sharpening exercise โ€” and it confirmed we were on solid ground.

The second is the anonymous-initiator handshake: the destination proves its identity, the caller stays unidentified until the callee chooses to accept. For setting up a call or a chat, that is exactly the right default โ€” it exposes the minimum to the relays in the middle. It is a primitive worth borrowing even when the underlying transport is something else entirely.

The third is announce-based discovery โ€” broadcasting a small signed announcement with a hop count and a strict bandwidth cap, letting it ripple outward, with no global directory anywhere. For finding peers nearby, it is a well-understood alternative to the usual lookup-table approaches, and a useful design point to measure ourselves against.

What we did differently (and why)

Reticulum is a fellow traveler, not a trunk we adopt, and the divergences are about our context rather than any flaw in theirs.

Our main transport assumes the internet is usually there, and is built for fast, direct connections over it. Reticulum is the opposite bet โ€” it assumes the internet may be gone and optimizes for slow, half-duplex, off-grid links. These are complementary slices of the same problem, not competitors. We see Reticulum as the right adapter for the grid-down and off-IP case โ€” a hive with one radio-equipped node staying alive through a full internet outage โ€” and not as a replacement for our everyday path. Our Wholeness axiom, which asks that a system depend on no external infrastructure for its core function, is precisely why this complementary lane matters to us at all: a community of ours should keep working when the network does not, and Reticulum shows that the lower layer can be made to honor that promise.

We also part ways on source-addressless envelopes as a global default. Reticulum eliminates source addresses everywhere. Our communities have named senders by design โ€” accountability inside a group is part of the point โ€” so we borrow the addressless pattern only for the genuinely privacy-sensitive moments, like setting up a call or dropping an anonymous note, rather than across the whole stack.

And our notion of identity is richer than a bare address hash: it carries a verifiable history, not just a key. A Reticulum-style crypto-address, for us, would be derived from that identity, never a replacement for it. That is our Honesty axiom showing through โ€” we want identity to be something you can check the provenance of, not merely a fingerprint that happens to match.

What we admire without reservation is how cleanly Reticulum satisfies the same principles we hold. No external dependency, no unencrypted link, no silent drop of an unsigned packet, and an addressing model held lightly โ€” announcements that expire and re-converge rather than accreting permanent state. That last point is quietly close to our own conviction that forgetting is a feature. We took the patterns; where we diverged it was our axioms choosing a different lane, with Reticulum's work standing fully on its own.


Written by AI agents from real project logs; owned and edited by Mujo.

โ† more in Vision   home โœฆ   all โ†’